Global water retention in forest canopy, litter, and soil layers and its controlling factors


 Forest ecosystems play a vital role in the earth’s hydrological process, and precipitation intercepted by forests accounts for more than a quarter of the water in the terrestrial hydrologic cycle. However, water retention in the three layers (canopy, litter, and soil) of forest ecosystems has not yet been thoroughly investigated on a global scale. Here, we investigate the global pattern of forest water retention capacity (WRC) and its controlling environmental factors based on 982 observations of 21 controlling factors in the three forest layers, mainly from 1990 to 2018. The results show that global WRC varies among the different forest types and climatic zones with a mean of 456.71 mm, while the average total water storage is 22,662.47 km3 in forest ecosystems. Climatic variables are the leading factors contributing to the variations in forest WRC, followed by forest structure factors, soil properties, terrain factors, and litter factors. This study advances our understanding of the mechanisms underlying large-scale variations in forest WRC in different climate zones and forest types. The findings demonstrate that controlling factors should be considered when developing policy for regions with important ecological functions. They also provide a benchmark to improve ecohydrological models for simulating global WRC.

can be used to simulate WRC globally; however, the simulation results typically lack large-scale 32 validation with observed data. The Gravity Recovery and Climate Experiment mission (GRACE) 33 tracked icesheet and glacier ablation, groundwater depletion, reservoir changes, surface water, and 34 soil moisture 10, 11, 12, 13, 14 . However, like the above models, the GRACE satellites could not further 35 divide the global WRC pattern into multiple forest layers (i.e., canopy, litter, and soil). 36 Although factors contributing to WRC have been individually evaluated, it is difficult to 37 comprehensively understand forest WRC without clarifying the relative effects of location, terrain, 38 climatic factors, forest structure, litter characteristics, and soil properties. Precipitation has been 39 reported as a major determinant of water storage 15 . The effects of changes in forest cover on 40 watersheds have been evaluated using paired experiments 16 . Vegetation cover and climate can have 41 offsetting or additive effects on changes in water resources in forested regions 4 . However, the effects 42 3 of various factors on water retention remain unclear because the individual effects are difficult to 43 distinguish when multiple factors interact with each other. In this study, we employed structural 44 equation modeling (SEM) to build a bridge between empirical and mechanical methods, quantify the 45 effects of the factors influencing WRC, and better understand the water cycle of forest ecosystems. 46 We collected 982 observations in 43 countries and regions from 254 peer-reviewed articles to 47 reveal the global pattern of WRC and its spatial variance in multiple forest layers (Fig. 1). This 48 approach allowed us to extend small-scale studies to the global scale and overcome the infeasibility 49 of large-scale WRC field measurements. 50 51 Fig. 1 A total of 982 observations from sites in the canopy, litter, and soil layers from a seven forest cover types 52 to intercept fog and cloud droplets 24 . MAP shows significant positive correlations with CIC in 111 different regions, and its correlations are stronger than those of other factors (Fig. 3), consistent with 112 previous results 25 . In contrast to MAP, evapotranspiration has a significant negative correlation with 113 global CIC, especially in the temperate and dry winter climate zones (Fig. 3). This may be related to 114 the large contribution of evapotranspiration to the basin water balance in this climate zone, where 115 evapotranspiration may exceed 90% of precipitation 24 . CIC is also positively correlated with tree 116 height, canopy density, and LAI in different regions (Fig. 3) to LAI and therefore is an essential factor in CIC ( Fig. 3 and Supplementary Fig. 2  is observed for CIC. This phenomenon can be explained by the fact that tropical forests capture more 131 precipitation in lower latitudes, resulting in larger CIC at lower latitudes 19, 32 . However, in the mid-132 9 high latitudes, litter mass and litter thickness increase with latitude, and more water is retained 133 because lower temperature leads to lower microbial activity and slower litter decomposition 33 . The 134 results of this study confirm that globally, LWHC is significantly positively correlated with elevation 135 and tree height, while it is significantly negatively correlated with MAT ( Fig. 3 and Supplementary 136 In summary, at a global scale, WRC is significantly positively correlated with elevation, slope, 146 MAT, MAP, tree height, LAI, stand density, canopy density, litter mass, and capillary porosity and 147 negatively correlated with latitude, evapotranspiration, interception rate, litter thickness, and non-148 capillary porosity ( Fig. 3 and Supplementary Fig. 6). Among the analyzed factors, the strongest 149 correlations with WRC are observed for evapotranspiration, canopy density, tree height, LAI, 150 interception rate, litter thickness, and non-capillary porosity (R 2 ≥ 0.80, P < 0.01; Fig. 4).   In this study, the effects of various factors were explicitly examined through a global synthesis 216 of multiple factors affecting water retention in the forest canopy, litter, and soil layers, thereby 217 extending the results from small-scale observational studies to a global scale. Our findings suggest 218 that both nature (e.g., terrain, forest structure, litter characteristics, and soil properties) and nature 219 drivers (e.g., climate change and land use change) have substantial effects on water retention in forest 220 ecosystems. Globally, four dominant factors (MAP, tree height, litter thickness, and soil porosity) 221 show synergic relationships with WRC, while evapotranspiration has an inhibitory effect on WRC. It 222 should be noted that the spatial and temporal distributions of data from the canopy, litter, and soil 223 layers used in this study are uneven among different climate zones and forest cover types (Fig. 1). 224 Nevertheless, the collected observational data cover 99.15% of forest types and 92.79% of climate 225 zones. Although the observational data span a large range of years, nearly two decades of observations 226 account for over three-quarters of all the data. Thus, improving the accuracy of global observational 227 data 10 is critical to refine and differentiate the different factors affecting water retention over time. 228 For example, unmanned aerial vehicle remote sensing can be used to control the relative errors in 229 measurements of forest structure factors (e.g., tree height, crown width, and DBH) 38 , improve the 230 efficiency of these measurements, and provide digitized data resources for forestry research. 231 With the worsening of global ecological issues, the human dimension of the forest-water nexus 232 has become more evident in recent years 39 . As the terrestrial human footprint continues to expand, 233 the amount of native forest free from severe damage from human activities is in precipitous decline 40 .  Table 1). Therefore, virgin tropical forests with high WRC values should be protected 246 and monitored. Given the broad distributions of savannas and woody savannas (Fig. 1a), these forest 247 types should also be valued for their WRC, and management policies should be formulated 248 accordingly. Furthermore, in these critical ecological regions for water retention, the main controlling 249 factors should be divided by climate zone and forest type. view holds that some regions should not restore ecosystems by afforestation because regions with 259 MAP values lower than 400 mm do not retain sufficient water to support trees 52, 53, 54 . In the future, 260 the effects of human activities on forest ecology and hydrology may be two sided, making the 261 attribution of forest water retention more complicated. However, forest-driven water cycles are poorly 262 integrated into global decision-making regarding land use and water management 24 . The next step is 263 to establish science-policy-practice scenarios to guide the management of global forest-water 264 resources and their related ecosystem services. 265

Literature data screening and extraction 267
We collected peer-reviewed literature for assessing water retention in forest ecosystems 268 ( Supplementary Fig. 9). First, we searched the Web of Science database using the following keywords:  (Fig. 1c). The data corresponded to 982 data sites, with 280 744 sites in the forest canopy layer, 120 in the litter layer, and 151 in the soil layer (Fig. 1a) (Fig. 1c). MODIS land cover data (MCD12Q1) 295 for seven forest cover types were used in this study (Figs. 1a and 1b). Soil data were from the soil 296 profile database of the World Soil Information Service. The Köppen-Geiger climate classification 297 map was obtained from www.gloh2o.org/koppen (Fig. 1b). The intact-forest-cover data are available 298 at http://www.intactforests.org 48 . 299

Integrated water storage capacity method 300
The integrated water storage capacity method was used to estimate WRC and TWS using 301 equations (1) and (2): 302 (2) 305 where A represents the patch area (km 2 ), and i represents a specific patch. CIC, LWHC, and SSC were 306 respectively calculated using equations (3)-(5): 307 where IR is the interception rate (%), LM is the litter mass (t/ha), WHR is the litter water-holding rate 312 (%), SD is the soil depth (mm), and NCP is the non-capillary porosity (%). 313 Seventy regions were identified according to the seven forest cover types and 10 climate zones, 314 and each region was ensured to have similar climatic and vegetation conditions ( Supplementary Fig.  315 8). For regions with no observable data, data from adjacent regions were used. We then calculated the 316 mean values and standard deviations (σ) of IR, LM, WHR, and NCP for each region. The mean values 317 were used to calculate WRC (Fig. 2), and the mean values ± σ were used to test the dispersion of 318 WRC ( Supplementary Fig. 1). 319

Polynomial curve fitting 320
Based on the least-squares method, we fitted the optimal curve of water retention capacity for a 321 potential controlling factor using functions (6) and (7): 322 p = polyfit (x, y, n), 325 where x, y represents the coordinates of scatter points to be fitted, n represents the power of 326 polynomial fitting, p represents the coefficient of polynomial fitting, and y1 represents the fitting 327 result using the p coefficient. 328

Statistical analysis 329
Due to the non-linear and non-continuous distributions of scatter points, spearman correlation 330 analysis was selected for further univariate analysis. The correlations were also evaluated for the 331 seven forest cover types and 10 climate zones. 332 Finally, the relationship between multiple factors and water retention capacity was tested by SEM. 333 SEM is based on a complicated regression relationship composed of multiple factors allowing the 334 effects of single factors on the population to be analyzed along with the mutual relationships between 335 single factors simultaneously. Factors can be divided into direct driving forces and indirect driving 336 forces that produce either positive or negative effects 49 . The total effect of a certain driving force on 337 the ecosystem is the sum of the direct and indirect effects. SEM has a variety of evaluation indexes: 338 a R 2 value close to 1 indicates that the observed variance in the model is well explained by the 339 considered controlling factor(s); a probability value (P) close to 1 indicates an excellent matching 340 20 effect between the model and the result; a ratio of chi-squared to freedom (x 2 /df) less than 3 and a 341 root-mean-square error of approximation less than 0.05 indicate that the model fits well. Using 982 342 observed data, we applied SEM to global forest regions and specified the factors controlling each 343 forest ecohydrological process. 344